Automatic Abnormal Cell Recognition Method Based on Image Splicing

2. The method of claim 1, wherein in 5), the cells are classified as 3 categories: normal, precancerous and cancer.

4. The method of claim 3, wherein in 5.1), the data augmentation is performed by a random combination of different data augmentation methods, comprising random clipping, horizontal flip, vertical flip, random light intensity transformation, and random angle rotation.

5. The method of claim 3, wherein in 5.2), all cell images in the same splice image are from the same slide image, and are spliced horizontally or vertically according to their dimensions to form a splice image with length being ROW×L and width being COLUMN×L; and a resolution of the splice image is increased while a number of channels thereof remains unchanged.

6. The method of claim 3, wherein in 5.3), the deep neural network model is added with two parameters α and β based on Residual Neural Network (ResNet) to control a resolution and a width of a network structure so as to adapt to different numbers of rows and columns possibly adopted under different image splicing strategies.

9. The method of claim 8, wherein in 5.1), the data augmentation is performed by a random combination of different data augmentation methods, comprising random clipping, horizontal flip, vertical flip, random light intensity transformation, and random angle rotation.

10. The method of claim 8, wherein in 5.2), all cell images in the same splice image are from the same slide image, and are spliced horizontally or vertically according to their dimensions to form a splice image with length being ROW×L and width being COLUMN×L; and a resolution of the splice image is increased while a number of channels thereof remains unchanged.

11. The method of claim 8, wherein in 5.3), the deep neural network model is added with two parameters α and β based on Residual Neural Network (ResNet) to control a resolution and a width of a network structure so as to adapt to different numbers of rows and columns possibly adopted under different image splicing strategies.

14. The method of claim 13, wherein in 5.1), the data augmentation is performed by a random combination of different data augmentation methods, comprising random clipping, horizontal flip, vertical flip, random light intensity transformation, and random angle rotation.

15. The method of claim 13, wherein in 5.2), all cell images in the same splice image are from the same slide image, and are spliced horizontally or vertically according to their dimensions to form a splice image with length being ROW×L and width being COLUMN×L; and a resolution of the splice image is increased while a number of channels thereof remains unchanged.

16. The method of claim 13, wherein in 5.3), the deep neural network model is added with two parameters α and β based on Residual Neural Network (ResNet) to control a resolution and a width of a network structure so as to adapt to different numbers of rows and columns possibly adopted under different image splicing strategies.